Escolar Documentos
Profissional Documentos
Cultura Documentos
(Kerry Plant)
Ampoule Inspection Machine
Eoin O’Sullivan
5th June 2014
Facility History
Location
AICL-Dublin
Plant
AICL-Kerry
Plant
Establishment September 1990
Established September 1990
AICL-Kerry Plant Official
Official Opening August
Opening August 1992
1992
Total Site Area 145,900 sqm.
TotalTotal
FloorSite Area 14,423
Space 146,400
sqm. sq.m.
Employees 294
ASTELLAS IRELAND CO., LTD
OLD NEW
Ampoule Visual Inspection
OLD NEW
Overview of Activities
Summary of Processing and Packaging Lines
Processing/Packaging Line Product(s) Processed/Packaged
• Machine throughput is
150 Ampoules per
minute
Collection of Sample Rejects
• Prior to implementing the project, live sample rejects from manual visual
inspection were collected over a two year period.
• Sample rejects were identified by applying their unique letter and number to
the tip of the ampoule using a thin black marker or by means of U.V. pen.
Theory of Operation
• The AIM has 22 separate cameras - each designed to detect specific types
of defects
• Particles
• Colour ring
• Scratches / cracks
• Tip defects
• Blue Dye
• Fill Volume
• Colour ring, scratches, cracks, tip, fill volume and blue dye defects are all
detected by absolute methods – i.e. measurement against set points
• The oscillating drive allows camera to move with ampoule when taking
pictures
– No stop / start action required with the machine for inspection – smooth ampoule
flow path
Pre - Spin
• Detection probability
– Characterisation of the particle detection process was driven by trying to
maximise detection probability.
– For particles this drove the design of the system with multiple cameras to
increase detection probability.
• The ideal solution would be 100% detection of all defects with no false
rejects thereby minimising the reject rate.
• However no vision process (manual or automatic) can give 100% detection
performance.
• Therefore in the interest of quality and safety we erred on the side of higher
defect detection performance and compromised with increased reject rate.
Ideal Balance
Reject Rate
Detection
Performance
Reject Rate vs. FRR
• Since the FAT / SAT the AIM sensitivity has been adjusted to minimise FRR.
• However getting real data on the FRR is difficult because
– 1) The AIM can see more defects than a human – so classifying what is real
and what is not is difficult.
– 2) The FRR has a random element to it – e.g. dust caused by environmental
conditions – the AIM sees the dust and captures this image but when checking
the ampoule later the dust is gone.
– 3) Some FRR is possibly still caused by variation in ampoule shape (e.g. base)
Most this has been tuned out now
Visual Inspection Approach
AIM Layout
Functional & Challenge Testing
• After the functional test, a Challenge test is run through the machine ensure
all cameras are functioning correctly.
• The Challenge test is made up of at least one type of each of the critical
rejects the AIM is capable of detecting. These rejects are from commercial
batches.
• The AIM must reject all ampoules in the challenge set before proceeding to
inspection of the batch.
Visual inspection of a batch
• Ampoules are loaded onto infeed belt and pass through all inspection
stations of the AIM.
• Reject particle ampoules are discharged to reject tray no. 1, reject cosmetic
/ tip ampoules are discharged into tray no. 2
• Forced rejects (if the machine is powered down during the inspection or if
an alarm occurs) are discharged into tray no. 3 for re-inspection on
completion of the main batch
AQL
• An AQL is from every batch processed through the AIM is manually visually
inspected.
Critical is any particle >200 μm & any fibre > 2000 μm,Tips, height, blue
dye, chip, liquid volume & cross contamination
Major is any particle <200 μm & any fibre < 2000 μm
Manually inspect
315 ampoules as
per FAP016
Are there
defects?
1 defect
0 defects
Challenge Solution
Not having a clear requirement of what the AIM List of defects put together & some ampoules
needed to reject & no measurement of particle sent off site for sizing.
sizes
Difficulty in detecting all defects types - cracks, Combination of backlight and reflective light as
particles, shape, colour well as colour and greyscale camera systems
Two fill volumes – 2mg & 5mg Upper and lower camera sets with minimum
changeover required
• If a defect falls out of solution (e.g. sticks to the upper wall of the ampoule)
then there is a likelihood of no detection.
• Incoming ampoules had a raised base causing false rejects, changes made
to window on base camera to accept these ampoules.
• False rejects were occurring during 2mg ampoule visual inspection due to a
reflection on the meniscus of the liquid - The two upper inspection window
positions were changed to “tilt” them in order to follow the meniscus slope.
• - A document was created detailing the breakdown of each recipe and its
corresponding current version number. There is a requirement in the MBR
to check the recipe on the AIM screen versus the document ensuring both
recipes match.
THANK YOU